Randy Neil Biostats 11: USMLE Step 1 - Kaplan-Meier Curve with Questions

USMLE purpose: Read Kaplan-Meier curves quickly: identify median survival, censored data, event drops, and group survival at a time point.

How to use this page

1. Learn what each part of the curve means before doing calculations.
2. Practice the “draw a horizontal or vertical line” method for median survival and survival at a time point.
3. Test yourself with the rapid recall and practice questions before opening the answer toggles.
4. Use the transcript only if you want Randy’s original wording.

One-liner

Kaplan-Meier curves estimate survival over time ; vertical drops are events, flat steps are event-free intervals, and censored patients remain accounted for without being counted as events.

Curve-reading workflow

Board Exam Buzzwords

Common traps

⚠️ Trap: Treating censoring as death.

Fix: Censoring means the event status is unknown after that point or the study ended before the event.

⚠️ Trap: Using the mean for skewed survival data.

Fix: Median survival is usually more useful and board-testable.

Anki-style rapid recall

Practice Questions

Question 1

A Kaplan-Meier curve drops sharply at year 4. What does the drop represent?

• A) Censoring
• B) Event occurrence
• C) Study enrollment
• D) Randomization

Question 2

Which group has better survival on a Kaplan-Meier curve?

• A) The curve with the steeper drop
• B) The curve that stays higher over time
• C) The curve with more censoring
• D) The curve that reaches 50% first

Quick Reference Summary

📌 Transcript note: The transcript is kept below for reference only. Use it if you want Randy’s original explanation after mastering the high-yield review above.

Full Transcript

Alright guys, so in this video we're going to talk about that Kaplan Myer curve. This is something that I've seen every now and then and essentially what I did is I went to the usemaly.org website and you know kind of pulled this the USMly content outline. And if you go on page 27 where it has the bios to statistics epidemiology, it literally tells you that the expectation is that you're going to meet your held accountable for understanding the Kaplan Myer curve. So in this video we talk about the several topics that are associated with that. I think we kind of hit it big stuff. But I will ask for your help on this if you come across any questions that are kind of on this Kaplan Myer curve content. If you went in mind sending them to me we can kind of use those as a base to kind of build from. And we can kind of take them kind of twist them and kind of make them our own. We just want to make sure we can see questions from different perspectives. If you find the video helpful, don't forget to subscribe, you know hit like all that kind of stuff. It does it does matter. It's something that we kind of track to make sure people are actually finding value and kind of in us doing this. So you know keep setting hard and hope like the video. All right guys. So here we go. Question number one says, Based on the above Kaplan Myer curve, what can we understand about the median survival time of group one versus group two during the study? Well, you know anytime again anytime you see like a stair stepping graphs, I want you to think this Kaplan Myer curve. Now there's a couple of principal takeaways from this. So anytime you see this, you get to think Kaplan Myer curve. Now the purpose of this thing, okay, is basically it's a way to estimate survival. Okay. It's a way to estimate survival. That's the whole purpose of this. The other thing that you need to understand is that when

they made this Kaplan Myer deal, they needed a way to incorporate what they call, and this is a keyword that you're going to see in this stuff is called censored data. And what that means is like, you know, when they do a study, and all of a sudden say, say somebody drops out or someone doesn't reach, doesn't have the occurrence or the event during that timeframe, then they have to have somehow account for this censored data. And that's what this Kaplan Myer thing does. It puts it kind of embeds it in there, and I'll show you that in just a second. So number one, you got to think stair step Kaplan Myer curve. It estimates survival. Okay. That's the purpose of it. It's going to have this stuff called censored data in there, and this censored data, just so you know, that can affect, okay, that can affect the curve. Okay. You know, the obvious, right? It maybe won't be as steep. It's a way the curve looks, you know, we'll get that in just a second as well. Okay. So back to the question. Well, again, based on the above Kaplan Myer curve, what can we understand about the median survival time of group one versus group two, or could be, you know, a treatment group versus a placebo group. Now they're going to identify this, but long story short, you got group one, which is going to be this highlighted one right here on top, and group two, the people who are in this one. Okay. And if you look on the y-axis, it's survival probability. So right here at 100, you know, they have an n%%, but they could just put it like, say, even decibels and stuff like that. But survivability at 100, that means this one is study began, everybody's alive, right? Of course. And then all of a sudden group one went this way, group two went this way. And so 100% of the people who are living here means 80% 60, 40, and 20% and if you're down here, that means well, everybody's dead. And then on the

x-axis, it's time. Okay. It's going to be time. And I just had a year in years, so 510, 15, 20, 25, 30, and 35 years. So this Kaplan, my curve is for this period of time. You know, say it was a one group received a treatment, second group didn't, and they followed them over a period of 35 years. We're going to determine in this between the survival rate, and then we got in this thing takes in consideration censored data. All right. So now here goes, is it a group one has a median survival score half of the patients, half of that of the patients treating group two. So they're saying group one has a median survival score half of this guy. Is it the median survival time of group one is approximately 15 years? Okay. Is it group one has overall reduced survival time compared to group two? Is it D, the median survival time for group two is eight years, or is it E, the median survival time for group two is half that of group one. All right. So a lot of stuff in here, but they'll long story short, it keeps asking about the median survival time. Remember what we did? You remember there was median, median, median mode. Mean is just the average, right? The median, like when you're driving down the road, that's just the middle, and then of course mode is most common. Okay. Well, the only one we're dealing here is median. So when talks about the median survival, well, here's survival. So what's the median? What's the middle? It's going to be, well, 50%. All right. It's the, that'd be the median. So if we were to draw that line all the way across, you know, it would cross group two right here and cross group one right there. So in group two, we would go down and so that median survival of group two is going to be, you know, let's just say looks around eight years. In group one, it's going to be the median survival is going to be, let's just say roughly looks like 2223 somewhere in

there. Okay. So back to our question. Group one has a median survival score half of the patients, a group two. So the same group one is half of group two. No. Yeah, it's time to close. Group one's median survival scores 22 years. So if you were a patient in the group one, and you took that medication or you do whatever that one was your median survival was going to be 22 years. If you were in this guy and took that medication or whatnot, your median survival score is only eight years. That's why you need to understand this whole capital and my or how to interpret it because you know, this is how they do some, you know, and read articles and such practice space to care evidence base care. I'm sorry. So group one. So it's not going to be this guy because that's opposite almost. Is it the median survival time of group one, which is a highlighted is approximately 15 years. No, 15 years would be right here. And so that's definitely not the median. Okay. Is it group one has overall reduced survival time compared to group two. Wait a second. Group one has an increased median survival compared to group two. So it's not reduced. It would be actually increased. So it's not him. Is it the median survival time of group two is eight years. Oh, yeah, here group two is eight. Well, kind of looks good. Like it. The median or is it you the median survival time for group two is half a group one. Now, it's group two is reduced, but it wouldn't be half. It was half to be more closer to 10 or 11. So, you know, it's it's not correct. Not bad, but it's not correct. The best answer. The only answer is going to be D. The median survival group two is eight years. So that's just how to interpret this this Meyer. Kaplan Meyer curve. It looks like stair steps. You're going to think Kaplan Meyer. You're going to think estimates survival. So if you're in this group, you know your median survival

was here. If you're in this group, it means for our own is there. And then you have to talk about censored data here in just a second. What that means is you know, censored data is essentially if someone, you know, if someone either just dropped out. Okay, dropped out of the study or they didn't have the event, you know, again, like if we were trying to see if someone had, you know, 10 people taking a drug had a mile card in in function. I was saying, one person in this group had the mile card in function. And so then then it goes down, and there's 90% left. And then another person, another person. But say somewhere in here, somebody dropped out. They had these things called these these little censored data. That would that you would have to incorporate and what that would do is it would make the stair step even longer. And that's just a mathematical way. So anytime you see a giant stair step. Compared relatively speaking, you know, you might consider that that is where censored data occurred. But for you right now, just know that censored data is part of this part of this deal. Okay. So the next one. It says, so same graph. It says, based on the above, Kaplamira curve, which group had a better 10 year 10 year survival prognosis. Okay, again, group one group two, they're asking that 10 years who had the better survival prognosis. Well, let's just say here's 10 years. And if we went all the way up, we can see that at 10 years group two this guy. Yeah, let's just say for simplicity, it was at you had a 40% survival. So if you took that drug or in that portion of the study at 10 years, you had a 40% chance of surviving. If you were in group one, you're looking at 80% roughly. Okay. So of course you'd like to act and I want to be in group one for this. So is a group one. Has a 10 year survival prognosis of 40%? Nope. Not group one. Group one was 80%. Okay. So we know

it's not him. Is it group two has a 10 year survival prognosis of 16? You know, that's not, it didn't make sense, right? It's not 16% and and 16, you know, that's that that's kind of saying right here has nothing to do with that's that 16 years. The survival group two would have been like say 10%, but it's just a play on numbers. So it's not that guy. Is it see group one has a 10 years survival prognosis of 80% group one 10 years 80% survival looking good right there is a group two greater than group one. No, now. Because the survival is much less, much less in this guy that it is in the group in this group. And you can see one one way I've seen a guy talk about it is pretend if you were like on a on a roller coaster or something like that. And the steeper the curve, right? You're more scared. So the steeper the curve you're thinking, okay, that's more dangerous not it, you know, for simplicity's sake. And so if it was a smaller curve, okay, it's easier ride not as dangerous. That's one way to look at it or you just understand that this is time you go up you find out where what year at five years, you know, the rate was 60% in group two. Okay, so the correct answer on that one, you're going to be see group one has a 10 years survival prognosis 80% capelin Meyer, okay, and now the last one. It says based on the above capelin Meyer curve, which of the falling time spans most likely had a censored patient. Now, you know, you to really understand this how the stair stepping occurs it's almost like this. You know, if you had in equals 10 in a study, okay, kind of totally separate right now. And so right here you started with a hundred percent everybody's in the study everybody's alive and say you start and then here's time. So as you start moving time, all of a sudden somebody has an event say it's a myocardian function or something, I don't know, it doesn't matter.

Someone has an event. So now, okay, one person at a 10 had the event. How many how much is left, okay, 90%. So I got to take 10% out, right, so this guy would go to 90. And then there should be, you know, basically nine sections left. So now I'm left with how many people in the study. So I'm going to go to 90, okay, and then I start moving with those nine people and then all of a sudden something happens right there. So two out of nine. Now that's a percent. Now that's, all right, I forget how much the how much that is. But then it goes down whatever two divided by nine is and then I should have nine sections. Okay, and then all of a sudden say I have two peak a couple people drop out. Okay, let's just say they fight out. Go away. They go away. They put in these things called censored data. So right here at the next event of someone who's still in the study, how many people were actually left. Let's think about that, right? How many people were left? One guy got out here because he had an event. Another person got out here. So now I'm left with eight. Then two people dropped out. So now I'm left with six people. So when this event happened, happened to one person, there were six left. And now I'm a sudden that's like 16%. So look, this drop here was 10%. Okay, and then two out of nine somewhere here. But now this is a pretty bit, you know, instead of it being one out of eight, because two people dropped out, it's one out of six. And so that's a lot bigger than it should be, right? So that drop is going to be slightly bigger per se. Okay, it's going to be slightly bigger. So in general terms, just think of it like that. And that means that chances are somebody dropped out of the study right there and then you go on and then, you know, either, either everybody dies or everybody has the event. Or, or the study ends, you know, ends at, well, ten years or something like

that. And then there's still some people that never had the event. Okay, that's the difference. That's a difference between something ending when it stair steps and ends like this. Or the study that ends like, you know, that. That means everybody had the event. This means some people didn't have the event and it ends just like that. Okay, anyways. The moral of me telling this is, you see, once I had it, one out of six and not one out of eight, it was a bigger drop right here. So based on the above, Kaplan Meyer curve, which of the following time spans most likely had a censored patient. Okay, is it a right here? B, C or D, and now you're thinking, all right, well, let's just go and eliminate something. And these look equal, right? Like you went down by roughly the same. Even though we know it's going to be a little bit different. There's no drastic changes on that. So I'm eliminating him. This, you know, just because the time is this long, do I know somebody left? Because the drop is not that much, right? It's a little bit more. Don't get me wrong. It's a little bit more than these guys. But it's not overwhelming, you know, it's not a giant drop. But when I look at B or A, I see, okay, there's a drop, they're pretty significant. Like, look at the difference here to here. Okay. And then here to here, well, we don't know how many people were in the study, but it looks like it could just be a proportional one. And nothing compared to. So the place where I think that perhaps there were some sensory patients or patients who just dropped out of the study and then no longer in the denominator, like they should have, I would say, you right to me, it's going to be right here, okay? Because let's just say, from here to here, it might have been out of, you know, again, I'm just going to say one, two, three, four, you know, five. It could have been, you know, one person out of

15, but then all of a sudden we had two or three people drop out. So the next one's going to be one out instead of being one out of 14, it's going to be, say, one out of 12 or one out of 10, and it's a bigger drop, okay? So that's kind of, it's just a way of doing it. It does a work every time, you know, I don't know. This stuff's, you know, if we can just understand that a censored patient is one part of the capital of myer, because they wanted to create a curve to where they could incorporate when people dropped out, can we still have it to where we can still do some type of chart to compare treatments of a treatment group, a placebo group or two treatment groups, to compare these even when peach patients drop out or, or if they never have the event, okay? If they never have the event, that's called a, like a right, a right censored or a censored patient. Okay? So the correct answer I'm putting there is, be as in boy. So, you know, where I'm getting this stuff guys is, you know, if you look, again, you have semi-emly outline, you know, I'm just going to try to go through here, you know, by a stats epidemiology, population health is kind of my thing. The survival analysis, the capital of myer curve is right there, they're basically telling you, they're telling you, you better be familiar with this. And so, if you really just do it that way and, you know, if you, if you really just understand that it's a way to estimate survival, and you're comparing the two, know the median piece and know how to interpret time, going up on these, I think it'll be just fine. Okay? I'd like to do you guys.